llms.txt git:20261008.d689213A

Immutable. This exact content never changes and is served at /api/v1/blob/5d5f9ac3a81095da.

# Sno Station

> Open-source software that turns the coding agents a developer already pays for (Claude Code,
> Codex, OpenClaw) into one squad on their own machine (Sno Station is the agents' workstation): shared encrypted memory, agent-to-agent
> messaging (Sno Reach), agent collaboration skills for handoff and cross-harness review, and a nightly loop that
> rewrites the agents' own skills with human approval. No daemon, no cloud required. Apache-2.0.

This repository has been public since 2026-09-18. Blocks in the README carry a "Last updated" date.

## Read first

- README.md — what it is, install status, how to use, the two hooks, benchmarks (below the fold)
- VISION.md — why it exists, the three reasons for two agents, "the workspace learns"
- docs/security.md — what is encrypted, where the key lives, what is and is not protected
- docs/sno-commands.md — every sno command, what it does and who runs it (the only command is sno)
- docs/readme/ — the README in 中文, 繁體中文, Deutsch, Español, Français, Русский, 한국어, 日本語

## Directory map

- packages/ — shared libraries (chunking, common-core, utils, embedder, observability,
  sqlite-crypto, content-sanitizer, memory = the memory engine + sidecar)
- apps/ — per-harness memory integrations (mem-claude, mem-codex, mem-claw, mem-hermes), Sno Reach (agent messaging),
  `sno heartbeat`, `sno report-time`, `sno subscription-quota-check` (programs run through `sno`)
- evals/ — one README + one formal result per benchmark (LoCoMo, Memora, R@5, Sno Memory Bench);
  no datasets, no raw runs
- docs/ — setup guides for the four memory integrations, images, evidence, translations, security page
- tests/apps/ — public-contract tests only

## Contributing

- CONTRIBUTING.md — DCO sign-off, one change per PR, which checks to run
- .github/ISSUE_TEMPLATE/ — bug, feature, design partner
- SECURITY.md — report vulnerabilities to security@sno.ai, never in a public issue

## Vocabulary

- squad — the category word: your agents working as one team (always lower-case)
- Dual Brain — two agents with different roles on a shared task, such as implementation and review; either agent can take either role
- Agent Squad — a larger team of agents, each with a clear responsibility
- Sno Reach — the messaging program, always written as two words (`sno reach <verb>`)
- the RSI loop — the nightly self-improvement loop (`sno rem-reflect`)
- Sno Station — this product; the human command is `sno`, the agent entry is saying
  "Sno onboard" inside any supported agent